Very nice! Amazing repair! Nice to see these getting saved.

I've seriously considered going down this road many times if I ever come across a PCI voodoo for my DOS rig super cheap being sold as parts. I have the tools and "skills" for such a repair, I just haven't had the luck. 😜

StriderTR wrote on 2025-03-30, 07:42:

Very nice! Amazing repair! Nice to see these getting saved.

I've seriously considered going down this road many times if I ever come across a PCI voodoo for my DOS rig super cheap being sold as parts. I have the tools and "skills" for such a repair, I just haven't had the luck. 😜

Thanks for the kind words. I think I got lucky. However, I’ve come across a problem…

Tomb Raider - Perfect
Descent 2 - Perfect
3DMark99 - Perfect
Quake 2 - Some texture issues, mainly in the menu console screen and some spotting during game play

Will post some pictures tomorrow. I’m using Fastvoodoo drivers which work fine in my Diamond cards. I doubt it’s a driver issue though. Need to do some more research.

It's great to see the results of your inspiring work! Unfortunately, there are no programs that allow you to test TMU. I carefully examine the textures in the demo Donut_glide2x and look for pins torn off from the contact pads under the TMU memory chips.

Postman5 wrote on 2025-03-30, 10:02:

It's great to see the results of your inspiring work! Unfortunately, there are no programs that allow you to test TMU. I carefully examine the textures in the demo Donut_glide2x and look for pins torn off from the contact pads under the TMU memory chips.

Here is what Quake 2 looks like in-game. You can hardly noticed the spots on the ground. Everything else in the game looks normal. You wouldn't really notice the issue until someone pointed it out.

I ran set SSTV2_NUM_TMUS=1 before running Quake 2 and got the following result. All texture issues gone when set to 1. Do you know which TMU is disabled?

I couldn't see any texture issues with the Donut demo? (sorry about the dust and wires on the screen. I got to sort that out 🤣)

I used an IR gun to work out which TMU gets disabled when you use SSTV2_NUM_TMUS=1. TMU0 got 5c hotter than TMU1 when running timedemo's on Quake 2. So TMU1 is disabled with that variable.

I tested all the pins again on TMU1 (the TMU with all the damage) with some tweezers for movement and BINGO found the following leg disconnected from the pad. I must have missed it on my first sweep.

105 TEX_DATA_1_12

So here is a recap of what had to be repaired on TMU1

Red = Disconnected leg from PCB
Yellow = Trace repair
Orange = Had to grind down chip to reveal chip traces

I suspect the main cause of the errors in MOJO was the disconnected leg on the FBI

66 FB_DATA_34

Quake 2 is working PERFECTLY now. I ran some other games and 3dMark99 a few more times. I'm happy to call this card 100% fixed.

Thank you for sharing this valuable repair information!

It seems like a common issue that these cards have are pins being ripped from their pads and/or simply being knocked loose. I think to make repairs quicker on these reflowing all pins around the TMUs and FBI chips would be a first step.

I wish I could get mine working. I've gone through my board and can verify all traces are in tact, all pins on FB and both TMU's checked and connected and no components missing or shorted but still no joy at getting the card working. Every once in a while it comes out and I go through more checking and testing but alas. One day it will work again.

ZenecadE wrote on 2025-04-09, 09:40:

I wish I could get mine working. I've gone through my board and can verify all traces are in tact, all pins on FB and both TMU's checked and connected and no components missing or shorted but still no joy at getting the card working. Every once in a while it comes out and I go through more checking and testing but alas. One day it will work again.

Highly recommend you making a post about your card. Describe everything you’ve done so far. Provide high resolution photos and provide mojo logs. Someone may give a hint as to what the problem may be 👍

Perhaps such a guide exists 😀

Just a little update (and a nugget of gold for someone else trying to fix their card).

After running Quake 2 for a while, the textures started going a bit wonky. Restarted Quake 2, textures are perfect but then go wonky again after a while. I thought maybe some additional legs were loose on TMU1, so I reflowed all the legs. To my dismay, the texture issue still persisted in Quake 2. No issues in 3dMark99. Anyway, I had a think about it. Why were the textures go from working perfectly to wonky after a while. If they start off being perfect, then maybe it's not a connectivity issue between components.

So I had an idea....I changed the memory clock from 93MHz to 90MHz using v2oc1400.zip

and now all texture issues are gone. My FPS in Quake 2 went from 52 to 51.7 so no big loss.

I found some nice time demos to stress your Voodoo 2 with Quake 2. They both give the TMU's a good work out.

Sounds like the 3dfx chips on the card would benefit from active cooling.

Spark wrote on 2025-04-12, 11:04:

Sounds like the 3dfx chips on the card would benefit from active cooling.

There is a 120mm fan blowing right on it. IR gun say TMU/FBI chips sit around 45c. I’m using FastVoodoo drivers so maybe they overclock the Voodoo by default.

zuldan wrote on 2025-04-12, 11:21:

I’m using FastVoodoo drivers so maybe they overclock the Voodoo by default.

I have no idea if they do.
In the 2.46 version, see FastVoodoo2_4.6/driver9x/V2-auto.inf : CmdAdd=set,"SST_GRXCLK=90"
But in FastVoodoo2_4.6/driver9x/voodoo2.inf : HKLM,"Software\3Dfx Interactive\Voodoo2\D3D","SSTV2_GRXCLK",,"93"

Fun fact : I never had any problem with the official 30202 drivers before, but with my Duron and K7T Pro 2 (KT133) Win98 freeze at boot with them.
Everything works with FastVoodoo 2.4.6.